• Aucun résultat trouvé

Investigating Alarm Fatigue with AlarmVR: A Virtual ICU for Clinical Alarm Research

N/A
N/A
Protected

Academic year: 2021

Partager "Investigating Alarm Fatigue with AlarmVR: A Virtual ICU for Clinical Alarm Research"

Copied!
7
0
0

Texte intégral

(1)

HAL Id: hal-03235942

https://hal.archives-ouvertes.fr/hal-03235942

Submitted on 26 May 2021

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

Investigating Alarm Fatigue with AlarmVR: A Virtual

ICU for Clinical Alarm Research

Anil Camci, Patricia Abbott, Deborah Rooney

To cite this version:

(2)

INVESTIGATING ALARM FATIGUE WITH ALARMVR: A VIRTUAL ICU

FOR CLINICAL ALARM RESEARCH

Anıl C

¸ amcı

1

Deborah M. Rooney

2

Patricia Abbott

3 1

School of Music, Theatre & Dance

2

Medical School

3

School of Nursing

University of Michigan, Ann Arbor, USA

1

acamci@umich.edu 2 dmrooney@med.umich.edu 3 pabbott@med.umich.edu

ABSTRACT

Alarm fatigue, wherein healthcare providers can become desensitized to alarm events due to extended exposure, is the root cause of one of the top health-technology hazards. However, investigating this phenomenon in clinical envi-ronments can be problematic due to privacy concerns and the risk of interfering with patient care. In this paper, we report the results of a preliminary survey conducted to un-derstand current alarm conditions in clinical environments across the University of Michigan. Furthermore, we dis-cuss our efforts to document such environments with 360o video and Ambisonic audio recordings. We then describe the implementation of a virtual ICU, namely AlarmVR, which is designed to facilitate research into clinical alarms and the auditory factors contributing to alarm fatigue. The proposed system enables healthcare professionals and au-dio practitioners to engage with alarm design to understand the acoustic parameter space of clinical alarms in terms of their perceived urgency. We detail the implementation of the system and outline possible use cases for it in research and instruction.

1. INTRODUCTION

Alarms are integral to clinical environments. Alarm fa-tigue, a condition where healthcare providers can become desensitized to alarm events due to extended exposure, is considered to underlie one of the top health-technology hazards [1, 2]. However, investigating this phenomenon in clinical settings can be problematic due to the risk of com-promising patients’ privacy or interfering with their care. To facilitate research into clinical alarms and the auditory factors that contribute to alarm fatigue, we designed Alar-mVR, which is a virtual ICU implemented in three phases. In the first phase, we used 3D panoramic video and Am-bisonic audio to document a real-life training ICU to es-tablish an immersive audiovisual benchmark for an alarm-heavy environment. In the second phase, we conducted a survey with healthcare professionals to probe the current alarm conditions across various critical care facilities at the University of Michigan. In the third phase, we developed AlarmVR based on the information gathered from the first two phases. The resulting system is a room-scale virtual ICU, where we can rapidly deploy a range of alarm

con-ditions and log users’ preferences, activities and reaction times.

In this paper, we provide details of our immersive doc-umentation of an actual ICU and report the results of our survey, discussing how these efforts informed the design of AlarmVR. We then give an overview of the AlarmVR sys-tem and discuss its interactive and immersive affordances. Finally, we discuss possible scenarios where this system can be utilized as an immersive simulation platform for in-struction and research in healthcare.

2. RELATED WORK

There is an extensive body of work that investigates the relationship between the acoustic properties of alarms and their perceived urgency. In one of the earlier studies in this area, Edworthy et al. carried out listening tests with psy-chology students and found fundamental frequency, speed, rhythm and melodic structure, among other parameters, to have consistent effects on the perceived urgency of an alarm sound [3]. Furthermore, they observed various inter-actions among parameters: For instance, changes in funda-mental frequency prompted different degrees of perceived urgency when the alarm onset was altered. In a following study, Haas and Edworthy further confirmed that increased pitch and loudness contribute to faster response times [4].

Previous studies have utilized surveys with healthcare professionals and medical trainees to identify root causes of alarm fatigue in clinical environments [5, 6]. The fac-tors that affect the audibility, identifiability and perceived urgency of alarms that are specific to clinical environments have also been extensively studied. In 2003, Mondor and Finley conducted a study on clinical alarms and found that the perceived urgency of these alarms were not propor-tional to the urgency of the clinical situations that they in-dicated [7]. In a recent study, Schlesinger et al. evaluated the effects of signal-to-noise ratio (i.e., that between alarm sounds and clinical background noise), and found that op-erator performance can still be maintained with alarms that are softer than the background noise [8]. In another recent study, Bennett et al. identified auditory icons (i.e. sounds that can function as metaphors for the events that they rep-resent) to improve the identifiability and discernibility of alarm sounds in clinical settings [6].

(3)

pitch, researchers have also investigated the musical prop-erties of alarms [9]. Drawing from music cognition re-search, Gillard and Schutz conducted experiments on the effects of melodic content on alarm identifiability, and found that alarms with musically distinctive features can improve alarm identification performance [10].

Recently, ¨Ozcan et al. have proposed a collaborative approach to clinical alarm design [11]. They argue that the actors operating in clinical settings, such as clinicians, pa-tients and visitors, should actively inform and participate in the design of alarm sounds before such sounds are de-ployed in such settings. Promoting a human-centered ap-proach to medical alarm design, the researchers argue that a co-participation of the stakeholders should take place in living lab environments, where alarms conditions naturally occur.

Following a similar approach, we utilize VR to create a virtual ICU, where auditory factors that contribute to alarm fatigue, and how alarms are perceived in general, can be investigated in virtual situ. The platform presented here, which is under ongoing development, allows us to incor-porate healthcare providers into the auditory study of med-ical alarms in a virtual context without having to interfere with patient care in actual clinical settings.

3. PRELIMINARY SURVEY

We conducted a survey to evaluate the current alarm con-ditions in various clinical facilities across the University of Michigan and to understand some of the acoustic factors that contribute to how healthcare professionals perceive and react to clinical alarms. The survey contained written, graphical, and ranked-choice questions as well as a listen-ing section that asked participants to compare hypothetical alarm sounds in terms of their perceived urgency.

3.1 Participants

19 participants completed the survey. The average age of the participants was 39.3 (SD=9.29). Among the partici-pants were 8 physicians, 8 registered nurses, 1 nursing stu-dent and 2 certified registered nurse anesthetists (CRNAs). None of the participants reported having impaired hearing. 3.2 Results and Discussion

3.2.1 Reactions to Nuisance Alarms

Alarms that are accurate yet clinically insignificant are called nuisance alarms [12]. These alarms indicate non-actionable conditions that can distract the providers from patient care. As seen in Fig. 1, when asked to rate the frequency of actions taken to minimize distractions caused by nuisance alarms, the participants most frequently se-lected“check patient” (M=5.11, SD=0.30) and “resolve the alarm” (M=4.84, SD=0.30), [5.00 aligning with Often (71-90%)]. In decreasing order of frequency, other actions in-cluded “ignore alarm” (M=3.21, SD=0.34), “request as-sistance” (M=2.89, SD=0.29), “turn down volume of de-vice” (M=2.84, SD=0.33), and “replace dede-vice” (M=2.68,

Figure 1. Ratings for actions taken to address nuisance alarms.

SD=1.26) [3.00 aligning with Not Often (11-40%]. Partici-pants reported “turn off device” (M=1.74, SD=0.18) as the least frequent action performed to minimize distractions, aligning with Very Rarely (≤10%).

3.2.2 Acoustic Factors Contributing to Annoyance In one of the questions, the participants were asked to rank auditory factors that contribute to the annoyance caused by alarm sounds. The choices presented in this question were derived from our interviews with healthcare professionals and included loudness, high-pitched or piercing qualities, repetitiveness and melody. The results of the participants’ ranking of these factors in terms of annoyance are shown in Fig. 2, with the pitch of the alarm sound ranked as the most likely to cause annoyance, followed by loudness, repeti-tiveness and melodic content.

In a follow up questions, the participants were asked to identify any other acoustic aspects of alarm sounds that they perceive to be annoying. While one participant iden-tified the inability to adjust the sounds as a factor, an-other participant described the similarities between differ-ent alarm sounds to be annoying. Other responses included “inability to talk over an alarm and” “loudness not being in line with the criticalness of the monitoring”. One partici-pant mentioned that some alarm sounds give them anxiety and that the acoustic properties of those alarms make them feel very irritated especially when busy. Another partici-pant reported annoyance with not being able to sustain si-lence amid a persistent alarm even when they know the alarm is false, for instance, when the monitor mistakes shivering for ventricular tachycardia.

(4)

3.2.3 Similarities between alarm sounds

Acoustic similarities between alarm sounds can be a cause of confusion in clinical environments. In the survey, the participants have identified several monitoring devices that gave out similar alarm sounds. For instance, participants likened the sound of the bed alarm to those of the venti-lator, the wound vacuum, and the sequential compression device. One participant mentioned that the sounds of the intra-aortic balloon pump are similar to that of the dialysis machine.

A common cause of confusion was the sound of the pulse oximeter, which was described to give out similar sounds under different conditions, for instance, when it reads 98% blood oxygen level and when it is not read-ing anythread-ing durread-ing an MRI. The sound of the same de-vice at a 95% blood oxygen level was likened to that of an electrocardiogram (ECG) machine without a pulse oxime-ter tone. The IV infusion pump was also characoxime-terized as putting out similar sounds across many different states (e.g., 0.9% normal saline vs benign fluid; vasopressor vs other life sustaining liquids). The alarm that indicates a change in ECG was also likened to the alarm that a temper-ature probe gives out when it has been removed. Other ad-visory alarms were also described to share similar acoustic properties with those that could indicate urgent conditions. 3.2.4 Acoustic Prominence vs Perceived Importance The participants were asked to place common medical monitoring devices on a 2-dimensional graph with the X axis indicating acoustic prominence (-1 representing low acoustic prominence and +1 representing high acoustic prominence) and the Y axis indicating perceived impor-tance (-1 representing low perceived imporimpor-tance and +1 representing high perceive importance). The following de-scription was provided to the participants:

Alarms can have varying degrees of acoustic prominence: some can be piercing whereas others can be soft. The acoustic prominence of an alarm can depend on its loudness, pitch, and rate among other factors. All these param-eters contribute to an alarm’s perceived impor-tance. A loud alarm can draw more attention and be perceived to be more important than a quiet alarm. However, some alarms can be acoustically prominent yet indicate conditions that do not require your immediate attention. For instance, the smoke detector in your home can go off while cooking even when there is no active fire. In this case, the alarm is acous-tically prominent despite the lack of an action-able situation of high importance.

An interactive graph allowed the participants to drag nodes that correspond to medical monitoring devices any-where on the graph. They were asked to base this evalua-tion on their overall experiences with these devices rather than on specific cases with patients.

Figure 3. Responses to the question where the participants moved nodes representing medical monitoring devices on a 2D-plane with X-axis indicating acoustic prominence and Y-axis indicating perceived importance. The blue circle around a node represents the standard deviation.

The average of the responses to this question are shown in Fig. 3 with the blue circles indicating standard devi-ation. Most devices were placed in the High Acoustic Prominence – Low Perceived Importance quadrant with the exception of the Sequential Compression Device and the Feeding Pump, which were characterized as display-ing low acoustic prominence and high perceived impor-tance. These results indicate an overall disparity between the acoustic prominence of clinical alarms and how impor-tant or actionable healthcare professionals perceive them to be.

3.2.5 Comparison of Hypothetical Alarm Sounds

In the listening portion of the survey, the participants were asked to listen to pairs of alarm sounds and pick the one that they perceived to be more urgent. These sounds were designed using our alarm sound prototyping tool, seen in Fig. 6, which we will further detail in the next section. Four of the seven questions were aimed at comparing variations in a single acoustic parameter (e.g., volume, rate, pitch), whereas the other 3 questions compared pairs where two parameters were changed in opposite direction at the same time (e.g., both an increase in pitch and a decrease in rate). The alarm pairings in individual questions were as fol-lows: high pitch vs. low pitch; low rate vs. high rate; high volume vs. low volume; high pitch, low volume vs. low pitch, high volume; low pitch, high rate vs. high pitch, low rate; high volume, low rate vs. low volume, high rate; no melodic variation vs. with melodic variation.

(5)

Figure 4. Participants’ comparison of perceived urgency between alarm pairs with different acoustic properties.

With questions where two parameters were changed in opposite direction, participants generally rated a low pitch, high volume alarm as indicating a higher level of perceived urgency than a high pitch, low volume alarm when the alarm rate was kept constant. This shows a tendency to-wards prioritizing the loudness of an alarm, as opposed to its pitch, when determining how urgent it is. With the ques-tion where a low pitch, high rate alarm was paired with a high pitch, low rate sound, participants generally ranked the increased rate to indicate a higher level of perceived urgency. This shows a similar tendency towards rate being perceived as a more prominent indicator of urgency when compared to pitch. Interestingly, in response to the ques-tion where a low volume, high rate alarm was paired with a high volume, low rate alarm, the participants were split evenly. While this result aligns with how loudness and rate were ranked against pitch, it also indicates that loudness and rate has comparable effects on perceived urgency.

4. IMMERSIVE DOCUMENTATION OF AN ICU To develop AlamrVR, we first performed an immersive documentation of one of the intensive care units in the Clinical Simulation Center at the University of Michigan. These units are fully functional medical facilities that are designed for training purposes. Since these settings are not used for actual patient care, we were able to setup video and audio systems to capture controlled alarm situations.

We used an Insta 360 Pro camera to capture a VR video of the ICU. This camera can record 3D panoramic videos at 8K resolution using its array of stereoscopic lenses. Ad-ditionally, we used a Sennheiser Ambeo 1st-order

Am-Figure 5. Immersive documentation of an ICU at the Uni-versity of Michigan’s Clinical Simulation Center.

Figure 6. Alarm designer software interface.

bisonic microphone to capture audio in synchronization with the video. After the 4-channel A-format recordings were converted to Ambisonic B-format, they were ex-ported to Adobe Premiere for spatial alignment with the video. The monitoring devices were also recorded individ-ually for reference with close microphone techniques.

The resulting video informed the design of AlarmVR as outlined below. A screenshot from this video can be seen in Fig. 5. The video is currently available on YouTube. 1 4.1 Alarm Designer

For the design of new alarm sounds, we developed the pro-totyping tool seen in Fig. 6 using the multimedia program-ming platform Max. This tool allows us to rapidly sketch alarm sounds based on 8 parameters: alarm rate, pitch sequence, fundamental frequency, waveform type, wave-folding amount, low-pass filtering, amplitude envelope and main volume. Through waveform mixing, wavefolding and filtering, a wide range of timbres with varying degrees of spectral complexity can be achieved. In addition to set-ting a fundamental frequency, the user can input a pitch sequence to add melodic variation to the alarm sound. Fi-nally, an amplitude envelope controller allows the user to arbitrarily draw contours that define the dynamic evolution of an alarm sound over time. This tool was used both to prototype the alarm sounds employed in our survey and to determine the parameters displayed to users in AlarmVR.

(6)

Figure 7. (Top) View of the virtual ICU in AlarmVR with medical monitoring devices. (Bottom) View of AlarmVR with monitoring devices replaced with abstract models.

5. ALARMVR

We developed AlarmVR as a virtual platform to investigate auditory issues related to alarm fatigue and clinical alarms in general. Research in this area is often abstracted from clinical environments due to the risk of interfering with pa-tient care. Although training spaces, such as the one we used for our immersive documentation, can facilitate stud-ies on clinical alarms, these are expensive facilitstud-ies that are not available at every institution. AlarmVR leverages modern VR systems to make research into clinical alarms immersive and accessible at the same time.

The virtual ICU seen in the top panel of Fig. 7 was mod-eled after the training ICU that was documented in the first phase of our study. The environment, interaction mechan-ics and logging system were implemented using the game engine Unity.2 The system supports both virtual naviga-tion with hand-held controllers and room-scale VR experi-ences, where the user can navigate the VR space physically within the pre-defined play area. It can be deployed to a va-riety of VR systems including the Oculus Rift and the HTC Vive.

The system allows the tester to administer varying com-binations of alarm sounds that are mapped to the virtual monitoring devices with spatial and directional sound us-ing Google Resonance Audio.3 In the virtual ICU, the user hears the alarms binaurally and can mute or solo them either with direct interactions with the virtual device or with remote interactions using a pointer mechanism. Each user interaction is logged with timestamps and can be played back for review.

2https://unity.com

3https://resonance-audio.github.io/

resonance-audio/

Figure 8. A user controlling the parameters of an alarm emitted from one of the abstract models in AlarmVR.

The system consists of a second state where the medi-cal models are replaced with abstract shapes as seen in the bottom panel of Fig. 7. This allows the tester to disasso-ciate hypothetical alarm designs from specific monitoring devices when it is necessary to test the perceived urgency of a sound regardless of the medical condition it may indi-cate. Alarm events are visualized with pulsating red dots as seen in the figure.

In a third state of the system, the monitoring devices (either with representative or abstract models) are over-laid with the user interface (UI) seen in Fig. 8. The UI allows the user to adjust the acoustic parameters of the alarm sound emitted by the corresponding device in real time. This state of the system allows the tester to prompt the user with various tasks that promote the active partici-pation of the user in designing an alarm sound.

AlarmVR is designed to serve both training and re-search applications. On the training side, our platform can be used to acclimate medical students to current alarm conditions in clinical settings. Instructors can easily de-ploy various alarm conditions, monitor student reactions, and log the results for further analysis. Since the system supports both room-scale and virtual navigation, it can be adapted to various use cases and simulation scenarios. On the research side, AlarmVR can be used both to evaluate existing alarm conditions and to design new alarms in an immersive context. The users can modify alarms in real-time and experience them in a spatial context individually or in various combinations.

Our first user evaluations with AlarmVR slated for April 2020 was postponed due to the COVID-19 pandemic. We hope to continue with this evaluation in Spring 2021 and report our findings in a future publication. In the mean-time, we will continue to establish AlarmVR as a virtual platform that can facilitate research into clinical alarms.

6. CONCLUSION

(7)

reported their experiences with the inconsistent pairings between the acoustic prominence and the perceived impor-tance of the alarm sounds emitted by monitoring devices that are commonly utilized in clinical settings. They also characterized alarms with high-pitch and piercing qualities as causing more annoyance than the loudness, repetitive-ness or the melodic properties of an alarm sound. While the increase in pitch, loudness and rate indicated a higher degree of urgency as expected, pairing inverse variations in these parameters yielded some interesting results with respect to the prioritization of loudness and rate. We out-lined our efforts to document alarm heavy environments for immersive simulation purposes. We then described our implementation of AlarmVR and our ongoing work on es-tablishing it as a platform that facilitates the active involve-ment of healthcare professionals in clinical alarm research.

7. ACKNOWLEDGEMENTS

This project is supported by a University of Michigan MCubed Grant.

8. REFERENCES

[1] Association for the Advancement of Medical Instru-mentation and others, “A siren call to action: priority issues from the medical device alarms summit,” Arling-ton, VA: Association for the Advancement of Medical Instrumentation, 2011.

[2] ECRI Institute, “Top 10 health technology hazards for 2020: Expert insights from health devices,” ECRI Ex-ecutive Brief, 2020.

[3] J. Edworthy, S. Loxley, and I. Dennis, “Improving au-ditory warning design: Relationship between warning sound parameters and perceived urgency,” Human fac-tors, vol. 33, no. 2, pp. 205–231, 1991.

[4] E. C. Haas and J. Edworthy, “Designing urgency into auditory warnings using pitch, speed and loudness,” Computing & Control Engineering Journal, vol. 7, no. 4, pp. 193–198, 1996.

[5] A. K. Sowan, A. F. Tarriela, T. M. Gomez, C. C. Reed, and K. M. Rapp, “Nurses’ perceptions and practices toward clinical alarms in a transplant cardiac intensive care unit: Exploring key issues leading to alarm fa-tigue,” JMIR human factors, vol. 2, no. 1, p. e3, 2015. [6] C. Bennett, R. Dudaryk, N. Crenshaw, J.

Edwor-thy, and R. McNeer, “Recommendation of new medi-cal alarms based on audibility, identifiability, and de-tectability in a randomized, simulation-based study,” Critical care medicine, vol. 47, no. 8, pp. 1050–1057, 2019.

[7] T. A. Mondor and G. A. Finley, “The perceived ur-gency of auditory warning alarms used in the hospital operating room is inappropriate,” Canadian journal of anaesthesia, vol. 50, no. 3, pp. 221–228, 2003.

[8] J. J. Schlesinger, S. H. Baum Miller, K. Nash, M. Bruce, D. Ashmead, M. S. Shotwell, J. R. Ed-worthy, M. T. Wallace, and M. B. Weinger, “Acoustic features of auditory medical alarms—an experimental study of alarm volume,” The Journal of the Acousti-cal Society of America, vol. 143, no. 6, pp. 3688–3697, 2018.

[9] E. ¨Ozcan, L. Frankel, and J. Stewart, “Uncommon mu-sic making: The functional roles of mumu-sic in design for healthcare,” Music and Medicine, vol. 11, no. 4, pp. 245–255, 2019.

[10] J. Gillard and M. Schutz, “Composing alarms: con-sidering the musical aspects of auditory alarm design,” Neurocase, vol. 22, no. 6, pp. 566–576, 2016.

[11] E. ¨Ozcan, D. Birdja, and J. Edworthy, “A holistic and collaborative approach to audible alarm design,” Biomedical instrumentation & technology, vol. 52, no. 6, pp. 422–432, 2018.

Références

Documents relatifs

The situation recognition is based on chronicle recognition where we propose to use the hybrid causal model of the system and the expertise to generate the pattern event sequences

This fact confirms the concept of the antipathetic structure sensitivity of the reaction: the fraction of atoms located on the faces of nanoparticles increases with particle

4 Conclusions and Possible Extensions Using a special reinforced urn process, we have built a Bayesian nonparametric alarm system to predict catastrophic events, such as

Zhou et al [ 39 ] hypothesised that AMPK might be an important molecular effector, as (1) metformin causes a reduction in cellular ATP:ADP ratios in hepatocytes [ 31 , 32 ]; and

In two patients (one with a body weight of 3.3 kg), percutanous PDA closure failed due to protrusion of the device into the aorta or left pulmonary artery; these infants suc-

According to the ANOVA, for this group roughness is also significant, and for sharpness only the higher level compared to the other two levels leads to a significant change in

The method proposed in the paper consists in automatically analyzing the alarm logs, by using pattern mining techniques, to detect frequent alarm patterns (step α in Fig. 1 b)

Other cases examine process- es at the regional level: the social attitudes to geneti- cally modified organisms in Catalan agriculture, the implementation of a Regional Strategy